aboutsummaryrefslogtreecommitdiffstats
path: root/docs/report/introduction/methodology_kvm_vms_vhost_user.rst
blob: 0a0ea1be1e700fdf552def7b769604365ea8cd79 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
KVM VMs vhost-user
------------------

FD.io CSIT performance lab is testing VPP vhost with KVM VMs using
following environment settings:

- Tests with varying Qemu virtio queue (a.k.a. vring) sizes: [vr1024] 1024
  descriptors to optimize for packet throughput.
- Tests with varying Linux :abbr:`CFS (Completely Fair Scheduler)`
  settings: [cfs] default settings, [cfsrr1] CFS RoundRobin(1) policy
  applied to all data plane threads handling test packet path including
  all VPP worker threads and all Qemu testpmd poll-mode threads.
- Resulting test cases are all combinations with [vr1024] and
  [cfs,cfsrr1] settings.
- Adjusted Linux kernel :abbr:`CFS (Completely Fair Scheduler)`
  scheduler policy for data plane threads used in CSIT is documented in
  `CSIT Performance Environment Tuning wiki
  <https://wiki.fd.io/view/CSIT/csit-perf-env-tuning-ubuntu1604>`_.
- The purpose is to verify performance impact (MRR and NDR/PDR
  throughput) and same test measurements repeatability, by making VPP
  and VM data plane threads less susceptible to other Linux OS system
  tasks hijacking CPU cores running those data plane threads.